EP3002455B1 - Method for determining the operating parameters of a wind power plant - Google Patents

Description

The subject relates to a method for determining operating parameters of a wind power plant, in which the operating parameters can be determined using optical methods while the wind power plant is in operation.

Operating parameters within the meaning of the subject matter are, in particular, those operating variables that can be changed during ongoing operation. The setting of operating parameters, in particular the blade pitch angle, is of essential importance for the operation of a wind power plant (also called wind energy plant or wind turbine). By means of the blade pitch angle, it is possible to vary the inflow area for the wind on the turbine's rotor blades. The blade pitch angle is regularly automatically adjusted to the current wind speed by a motor, so that the electrical power generated is adjusted to the respective generator and/or the rotor rotates at a speed that is as uniform as possible. The yaw angle, i.e. the orientation of the wind turbine in the wind, can also be a relevant operating parameter. In addition, movements, in particular vibrations of the tower, the nacelle or the spinner can be relevant operating parameters.

If the operating parameters are not right, the conversion of wind energy into electrical energy can lead to reduced yields. This means that the system generates less electrical energy from the wind energy than would be possible with an optimal setting. Incorrect adjustments, in particular in the case of an incorrect yaw angle or an incorrect blade pitch angle of at least one rotor blade, can result in increased dynamic loads on the components of the wind turbine. In particular, the rotor blades, the tower, the nacelle, the drive train and the foundations are exposed to high dynamic loads if they are incorrectly adjusted. This leads to damage and in particular to a reduced service life of the wind turbine. In the Example of the blade pitch angle, it is of particular importance that this is the same on all rotor blades, usually between two and four rotor blades on a system, especially during operation, i.e. under wind load, but at least the same with a small tolerance. If the blade pitch angles of the various rotor blades are different from one another, the wind load per rotor blade is different and imbalances can occur in the rotor and in the drive train. The setting of the blade pitch can be done separately for each rotor blade, or the rotor blades are operated with the same blade pitch settings. In both cases, however, the setting parameter for the blade pitch angle does not mean that the rotor blade actually has this angle under wind load.

Imbalances can also lead to movements of the wind turbine, which also mean a reduced yield on the one hand and increased wear on the other. If vibration is discussed below, this preferably means not just a harmonic vibration, but any movement.

In addition, it is to be expected that the noise emissions from such a wind turbine will increase if the settings are incorrect. This is particularly relevant in view of the fact that such systems are increasingly being set up in the vicinity of residential areas.

It is currently possible to measure the blade pitch using sensors inside the wind turbines. However, this is problematic insofar as the sensors only record and output the actual angle of attack if they are correctly calibrated. Otherwise it is also possible to measure the blade angle of attack. However, nowadays this is only done when the wind turbines are stationary. On the one hand, this has the disadvantage that monitoring the blade angle of attack leads to a loss of production of the wind turbines. On the other hand, it is not the blade pitch angle that is relevant for the above-mentioned problems when the wind turbine is in operation, but when it is at rest. In particular, the blade geometry changes depending on the wind speed and speed. When stationary, a rotor blade has one Wind power plant most likely has a different geometry than in operation. Deflections and torsions of the rotor blade come into play here, which play no role when the system is at rest.

From the DE 10 2013 110 898 A1 a method is known that uses thermal signatures of rotor blades to determine the condition of the wind turbine.

The EP 2 369 176 A1 and the DE 10 2008 031484 A1 describe a method for determining a pitch angle of a wind turbine.

The FR 2 882 404 describes a method for measuring a rotor blade of a wind turbine.

From the German patent application DE 10 2008 053 928 A1 the monitoring of a wind turbine at a standstill using drones is known. Flying drones make it possible to carry out an optical inspection of the wind turbine at a short distance from the rotor blades.

However, this detection is extremely complex and due to the movement of the drones, incorrect measurements can occur.

For this reason, the subject matter was based on the task of providing a method for determining operating parameters of a wind power plant, which method supplies reliable data on the operating parameters during operation.

Objectively, this object is achieved by a method according to claim 1 .

It has been recognized that operating parameters can be recorded during operation by means of photographic and/or videographic analysis of the wind turbine. Furthermore, it has been recognized that ground-based detection can be quite satisfactory measurement results. In particular, the image acquisition can take place free of natural oscillations of the acquisition means. However, the image acquisition can also be possible, for example, from the hand and be affected by natural vibration. The detection means can be mobile and/or stationary (stationary). The detection means can be a video camera.

With the help of the present method, it is possible to record the blade pitch angle of at least one of the rotor blades in a particularly reliable manner. In this case, it is preferable for the images to be recorded approximately parallel to the axis of the tower. It is proposed that the optical axis of the image acquisition runs approximately parallel or at an acute angle to a tower axis of the wind turbine. This enables ground-based detection of the rotor blades. Furthermore, the rotor blades, in particular both during a zero crossing, in particular when the longitudinal axis of the rotor blade runs approximately parallel to the tower axis and these point towards the tower or ground, and during a horizontal position, in particular when the longitudinal axis of the rotor blade is approximately 90° to the tower axis pivoted are detected. In particular, the rotor blades are detected in the horizontal position that immediately follows the zero crossing.

Both recordings are possible by setting the optical axis parallel or at an acute angle to the tower axis. In this case, in a sequence of recorded images, the same image section (possibly also the entire image in each case) is viewed in each case. In this sequence of images, the rotor blades are recorded in two different positions (at two different times), namely in the zero crossing and in the horizontal position. The gondola and the spinner can also be captured in the same image section. By acquiring this information about the spinner and the rotor blades, which is relevant for the measurement, a blade pitch angle of at least one of the rotor blades during operation, ie while the turbine is rotating, can be acquired particularly accurately.

In particular, the inventors have found that the blade pitch angle can be measured particularly well when the rotor blade is in the horizontal position. This angular position can be detected particularly well when the optical axis of the image acquisition runs essentially parallel or at an acute angle to the tower axis. The recording from below makes it possible to measure at least one rotor blade during operation, with operation-related deflections and twisting of the rotor blade being taken into account. The ground-based measurement ensures that the recording device is not subject to any or negligible natural vibration. In addition, climbing the wind turbine is not necessary. In addition, the measurement can be carried out during operation, so that production downtimes are avoided. In addition, the measurement technology, in particular the videographic recording, can be used for a plurality of wind turbines and no additional measurement technology is required on or in the wind turbine itself.

In particular, the images are captured videographically from the ground. The images are then subjected to image processing, in which individual images or image sections that are relevant for determining the operating parameter are detected in a chronological sequence and then evaluated.

First of all, it is necessary to remove the vibration of the wind turbine tower from the evaluation. For this it is necessary to first detect the vibration of the tower. For this reason, it is proposed that at least parts of a nacelle and/or a spinner of the wind turbine and at least parts of at least one rotor blade of the wind turbine be captured in the field of view of the image acquisition. In the area of the field of view of the image acquisition, a characteristic point of the wind turbine, eg the spinner or the nacelle, is preferably to be identified, the position of which in the image is independent of the rotational movement, in particular at a fixed distance from the rotor. In the case of a spinner with a "flat nose", for example, these could be the two corner points. In particular, a spinner tip of the spinner can be detected as a reference point. The spinner tip is a well-suited reference point for determining the vibration of the wind turbine tower. rest the Image acquisition is stationary relative to the tower and the wind turbine, so the characteristic point moves depending on the oscillation of the tower. The vibration of the tower and/or the nacelle of the wind turbine can be determined by evaluating the position of the characteristic point in a temporal image sequence.

In order to determine the vibration, it is proposed that moving images of the wind turbine be captured. These moving images represent the wind turbine at different points in time. By evaluating the individual images, it is possible to evaluate the different conditions of the wind turbine at the respective points in time, in particular the reference point, the spinner, the spinner tip, the nacelle and/or the rotor blade.

In order to normalize the recorded images, ie to hide the movements of the tower and/or the spinner and/or the nacelle caused by the mechanical load on the wind turbine for further evaluation, an initial characteristic point is first determined in an image. This initial landmark is a point of the image that represents part of the wind turbine. This part of the wind turbine can preferably be recognized in each of the individual images of the sequence of recorded images that are used for the evaluation. This characteristic point is in particular a point in the images whose position is independent of the actual rotation of the rotor blades, in particular whose position is rigid with respect to the axis of rotation. In particular, this can be the spinner tip. Since initially only the vibration of the tower of the wind power plant is to be eliminated, the initial characteristic point is in particular a pixel of the wind power plant whose image position does not change between the individual images due to the rotation of the rotor blades. The rotor blades are constantly changing their position, so that the vibration of the tower cannot be determined from this. On the other hand, the position of the nacelle and in particular the spinner as well as the spinner tip are not influenced by the rotation during the rotation of the rotor blades. The characteristic point is preferably determined on these components. Only a swing of the tower leads to a change in the position of the nacelle, the spinner and the spinner tip. The initial characteristic point can preferably be determined manually in a single image.

In a daylight shot taken from the ground along the tower axis, the nacelle, spinner and rotor blades usually appear in very high contrast against the sky. This property can be exploited by performing edge detection in at least one of the captured images. This edge detection is particularly easy when the contrast between the wind turbine and the sky is very high. With the help of edge detection, the outer contour of the rotor blades as well as the spinner and the nacelle can be recorded. An edge detection can e.g. be carried out with a Sobel edge detector. The frequency spectrum of the images can also be used. In the frequency spectrum, a high-frequency area can represent a change in contrast, and therefore an edge. Appropriate high-pass filtering can be used to identify the areas and pixels that can represent edges. These can in turn be displayed in different colors (or grey) in the local area.

A change in the pixel values relative to a background is regularly relevant for edge detection. In order to always enable good edge detection under different light conditions, it is proposed that a reference value of a background parameter is determined. It is possible to select the background parameter from a contrast value, a color value, a gray value, a value of a color channel or the like, in particular manually.

Based on the selection of the background parameter, a reference value is then determined for this. For this purpose, a background area can be selected in at least one image. A background area is usually an area in which preferably only the background can be seen and not the spinner or other parts of the turbine. In particular, this area can be determined by the user in at least one image.

After the range has been determined, the reference value of the background parameter can be calculated. Here, for example, an average of all pixel values in the selected area can be calculated for the selected background parameter. This value can then be used as a reference value.

Edge detection can then be carried out. A further image area can be determined for this purpose, in particular manually. The spinner and preferably other parts of the turbine can be seen in this image area. The initial characteristic point is preferably also present in this image area. A deviation of a pixel value can be determined with the aid of the reference value. In particular, a deviation from a background pixel can be detected. With the help of the reference values, the edge of the spinner can be detected.

A curve shape is approximated at a detected edge. In particular, a second or third order curve can be approximated at the edges. Preferably, a parabola or some other conic section can be approximated at the detected edge. To do this, the pixels are determined that represent the edge with the highest probability. For example, if a picture element (pixel) exceeds a limit value in the frequency range, this pixel can be assigned to an edge. Any contiguous pixel with a frequency value above the threshold can be evaluated as an edge.

For example, curve parameters, in particular parabola parameters, can be determined during the approximation. The curve preferably does not change, particularly with regard to its parameters, between the images. However, it can also happen that the shape of the edge changes between the images. This can happen, for example, as a result of mechanical stress and compression. It can also be possible to separately determine curve parameters for several, in particular all, individual images. A change in the curve parameters can be used to evaluate the mechanical load on the spinner and/or the nacelle as operating parameters. However, a set of curve parameters can preferably also be determined in an image and used for further, preferably all, images. Preferably, the shape of a recognized edge does not change, only the position of the edge within the image shifts.

Preferably, the pixels are spatially adjacent along an edge. Brightness values can also be assigned to the pixels depending on their spatial frequency. For example, a brightness value of a pixel can also be determined after edge detection. Starting with one pixel, for example, neighboring pixels can be evaluated with regard to their brightness value and the brightness distribution can be used to determine where the edge is most likely to be located. The detected pixels are used to approximate the shape of the edge. This means that the course of the curve is calculated in such a way that it approximates the edges actually recognized in the image.

In the case of edge detection, the image is preferably displayed in shades of gray and the edges, in particular high-frequency components within the image, are displayed brighter than areas in which no edges are detected.

A reference profile of at least one known edge can also be used for edge detection. Here it is possible to store reference curves of spinner edges of known turbines in a database. In particular, such reference curves can be stored for specific types of wind turbines. The reference curves can be calculated from images of reference turbines. For example, these images can be captured immediately after installation of the turbine. The images can also be captured in very good lighting conditions. It is also possible to derive the reference curves from the construction data of a wind turbine. The reference curves can be stored in a database.

The edge initially detected in the image is compared, for example, with at least two reference curves from a database. The reference profile that has the greatest similarity to the detected edge, for example the lowest Deviation (e.g. in total) of the pixels in their position from each other can be saved as an edge of the current turbine.

The use of reference gradients helps in particular with edge detection when the lighting conditions in the captured images are not optimal. In particular in the case of shadowing of parts of the spinner, the edge can be approximated by extrapolation together with the reference profile. In particular, in the sequence of images, those images in which the edge cannot be reliably identified due to various disturbances can be approximated using the reference profile.

In the case of deviations from the reference course due to the detected edge, a local gradient of the deviation can also be determined, for example. If the deviation is rather abrupt, i.e. with a large gradient, a message can also be issued. This information can be used, for example, to identify dents, dents or other damage to the spinner.

The curve of the detected edge can also be approximated using an n-th degree equation. In this case, it is possible for a selection of parameters of the equation of the nth degree to be iterated, and for the equation to be selected which best matches the recognized edge. In particular, an equation of the 4th degree or also of the 2nd degree can be used. The user can be offered the option of selecting which equation or which degree is to be used.

Once the curve of the detected edge has been approximated, information about the curve is stored in memory. This can either be information about the reference curve or information about the equation with which the edge was approximated.

After the curve has been approximated by the recognized edge, the characteristic point is determined in the approximated curve. Ie that the point of the curve which lies closest to the initial characteristic point, is determined as the characteristic point. A manual determination of the initial characteristic point is thus optimized in that it is then shifted to an approximate characteristic point by approximating the curve through the edge. In this case, for example, a local or absolute maximum or minimum of the approximated curve can be determined and the initial characteristic point can be shifted to this point. The initial characteristic point can then be a calculated characteristic point. The manually determined characteristic point can be used to first identify the edge whose course is then to be approximated.

It is possible that the characteristic point is shifted to a zero point of the approximated curve. The zero point of the curve usually represents the top of the spinner. Particularly in the case of a parabola, the course of the parabola can correspond to the course of the spinner and the zero point of the parabola corresponds to the tip of the spinner. The zero point of the curve can be calculated.

The initial characteristic point can either be the manually determined characteristic point or the characteristic point shifted by approximating the curve (this characteristic point can then be regarded as the initial characteristic point). The initial characteristic point is initially determined in a first image of a sequence of images that are used for the subsequent evaluation. In particular, a first image is determined, based on which the subsequent images for determining the operating parameter are to be determined.

Then, starting from the first image, edge detection and curve approximation can be carried out for each individual image. A characteristic point can be determined in chronologically consecutive images, preferably in all chronologically consecutive images of a sequence. This respective characteristic point of an image is viewed with the initial characteristic point of the first image with regard to its spatial position within the image, in particular its X and Y coordinates within the image. A shift in the position of the characteristic point from initial characteristic point means a swinging of a tower and/or the nacelle. The deviation of the position of the characteristic point from the position of the initial characteristic point is preferably determined for each image in a sequence. This deviation of the position of the characteristic point of an image from the position of the initial characteristic point represents the deflection of the tower and/or the nacelle relative to the position of the tower and/or the nacelle at the time of the image of the initial characteristic point.

When evaluating the individual images, it can be useful to first determine the position and alignment of the edge for each of the individual images. In this case, the previously stored approximated curve can be accessed first.

A translation and/or a rotation of the spinner, for example, can be detected with the aid of the position detection of the characteristic point in the sequence of images. If a translation is detected, a vibration analysis can be carried out using frequency analysis, for example. An analysis of the movement of the characteristic point can also provide information on possible rotor imbalances in the turbine.

With the help of determining the spatial deviation of the characteristic points, the operating parameter of the movement of the tower at the time of measurement can be determined. The shift in the characteristic points can be plotted in a diagram and allows conclusions to be drawn about the vibration behavior of the wind turbine, particularly the tower.

When recognizing the position of the characteristic point and then evaluating the movement of the point, it is necessary to distinguish between a translation and a rotation.

In a first method, for each image, as described above, an edge can be fitted by mathematical fitting to a 2nd or 4th degree curve. It has been recognized that a comparison of the detected curve to a curve of the same degree with only even coefficients is useful for an evaluation of the translation and the rotation is possible. In particular, a curve from a polynomial with only even coefficients can be assumed to be axisymmetric. Assuming that a spinner is structurally also axisymmetric, it can be concluded that if the approximated curve only has even coefficients, that there is only a translation. The prerequisite for this is, of course, that the reference system, i.e. the image axes, is not shifted between the images. It is also useful if a preferably vertical image axis runs at least parallel, preferably through the axis of symmetry of the spinner. Any deviation of the detected edge from a polynomial with only even coefficients can be evaluated as a rotation.

The axis of symmetry of the detected curve can also be determined. In this case, the original curve can first be compared with the currently recognized curve. Assuming that the shape of the edge of the spinner does not change, the slope of the symmetry axis of the current curve can be determined from the original curve.

The inclination of the curve, ie in particular the inclination to the y-axis, can also be determined by evaluating the current curve. First, on both sides of the characteristic point, which in particular represents the apex of the curve, two points along the curve can be determined at an equal distance along the curve from the characteristic point. If these points are parallel to the horizontal image axis, there is only one translation. However, if the points are not parallel to the image axis, a rotation can be concluded. The rotation depends on the path integral between the respective points and a section through the curve parallel to the horizontal image axis, i.e. e.g. the path in pixels that must be covered by each point along the curve so that the shifted points in the image are at the same height are, i.e. have the same y-coordinate.

The detected movement can then be compensated for further evaluation of the images. For this purpose, all pixels of a single image, preferably each frame of a sequence is shifted depending on the determined deviation of the position of the characteristic point from the position of the initial characteristic point. In particular, the pixels are shifted exactly in opposition to the determined positional deviation of the characteristic point from the initial characteristic point. For example, in one image it has been determined that the spinner tip has shifted ten pixels to the right and five pixels up within the image. To compensate, the image or the pixels of the image are shifted ten pixels to the left and five pixels down. Then the characteristic point of this frame lies directly on the initial characteristic point. If this is done for each frame, the resulting sequence of frames is independent of the detected vibration. The characteristic point is in the same spatial position within the image in each individual image.

In the compensation, the rotation can also be compensated. The entire image is then rotated against the detected rotation of the spinner.

It is then possible, for example, to determine the period of rotation and/or the speed of the wind wheel. For this purpose it is possible, for example, to determine the zero crossing of each individual rotor blade. A graphic evaluation can also be carried out for this. In particular, it is possible to evaluate a mean value of the pixels contained in a specific image section, e.g. a mean value of the color values of the pixels, in particular a mean value of the gray values of the pixels. To determine a zero crossing, for example, that image detail can be defined that includes the area directly below the nacelle. The optical axis runs from the recording means along the tower to the nacelle. At the time of a zero crossing, a rotor blade crosses this optical axis. In particular, the rotor blade is essentially parallel to the tower axis at zero crossing and points in the direction of the ground.

It goes without saying that it is not always possible to determine a single image in which a rotor blade is exactly at the zero crossing. Rather, the goal is to capture that frame determine in which the rotor blade is as close as possible to the zero crossing. For this purpose, the image section is evaluated with regard to an average value of the pixels contained in the image section, eg an average value of the gray values, an average value of color change or the like. If the image section was selected below the nacelle, its mean value changes whenever a rotor blade is in this area and crosses the optical axis. A sheet throughput image can be determined by determining the image whose mean value falls below or exceeds a limit value. It is also possible to compare the mean values of the image sections of the individual images with one another. Those images in which the mean values reach at least a local maximum are determined as sheet passage images.

As already mentioned, the image section is in the image area that shows the gondola from below. In particular, the image section can be determined in an area between the spinner and the tower of the wind turbine.

A zero crossing of a sheet can also be detected by evaluating the position of the tips of the other sheets. With a recording in which not only the spinner, but also the rotor blades can be recognized, an evaluation of the position of the blades is possible. In particular, it can be detected, for example, in which images two blade tips have the same or at least similar distance to the rotor root, in particular to the axis of symmetry of the spinner. For example, if one blade is in a vertical position, the other two blades each point 60° outwards. In this case, the tips of the blades are equidistant from the tip of the spinner. The image can be searched for in which the difference in distances between each blade tip and the spinner tip is minimal. This image can be determined for each pass.

The advantage of this detection of the zero crossing can be that in the zero crossing one sheet points upwards and the other two outwards, so that the spinner tip cannot be covered by any sheets.

A wind turbine usually has more than one rotor blade, so that several zero crossings are detected during a complete revolution. The number of rotor blades in a wind turbine is known. If the time and/or the number of images that lie between two blade passage images is determined and the number of rotor blades is known, the period of rotation of the wind turbine can be determined. In particular, each image can be assigned a time stamp. Thus, the time between two images can be determined by evaluating the time stamp. The frame rate can also be known and the duration of each frame can be determined from the frame rate. This can then also be used to determine the time between two images.

In particular, a period of a complete rotation of the rotor blades is determined as a function of the specific time and/or the specific number of blade passage images. The time or the number of images between two blade passage images multiplied by the number of rotor blades gives the period of the wind turbine. This period can also be determined for each full revolution and entered in a diagram. The mean values determined can also be plotted in a diagram. Local maxima of the mean values can be determined in a zero crossing. With each zero crossing, the period can be re-determined by multiplying the distance or time of the images between the current and previous blade passage images by the number of rotor blades.

An image can also be determined in which the rotor blade assumes a horizontal position. This image can be determined from, for example, a leaf passage image, knowing the period. It is also possible, for example, to determine that image in which a tip of the rotor blade is at a maximum distance (in the image plane) from the blade root, in particular the axis of rotation of the wind turbine. The image can also be determined, for example, in which two rotor blades have an axis of symmetry along the axis of the tower. Then the rotor blades are in a so-called Y position, for example.

Based on this image and in particular the knowledge of the period or speed, the image with the horizontal position can be determined.

As mentioned, it is possible to determine the image in which the rotor blade is in a nearly horizontal position, for example starting from a blade pass image. A rotor blade is in a horizontal position, for example, exactly a quarter of a period after the blade pass image. If the period is four seconds, for example, i.e. the rotor blade rotates exactly once in four seconds around the axis of the drive train, then a rotor blade that has passed through zero is in a first horizontal position exactly after one second. This is the horizontal position following the zero crossing, which can be taken into account for the following consideration. In the case of a recording with fifty images per second, the fiftieth image could be determined after the blade passage image, in which the rotor blade is in a horizontal position. Such an individual image determined on the basis of the period can be determined as a measurement image. The image in the second horizontal position, which is reached after 3/4 of a period starting from the zero crossing, can also be used for the evaluation as a measurement image. The speed and/or period can be determined continuously and a determination can thus be made based on a current speed and/or period. It can also be assumed that a speed and/or period is constant over a period under consideration. In the measurement image after 1/4 of a period, the rotor blade is recorded in a horizontal position from below.

Also, a horizontal position can be calculated from the Y-pose image. For example, a rotor blade may be in the first horizontal position in the image if, starting from the Y-position image, the image captured 1/12 of the period before the Y-position image is selected.

In the horizontal position of a blade, for example, the distance between the tip of the blade and the tip of the spinner can be determined. Here, for example, a distance along the x-axis and a distance along the y-axis of the blade tip from the spinner tip to be determined. If this distance is determined for each sheet, a difference between the sheets, for example with regard to their deflection, can be determined. This allows, for example, a conclusion to be drawn about material wear on the respective blades and an incorrect yaw angle. Also, an angle between an axis perpendicular to the axis of symmetry of the spinner and an axis through the spinner tip and the blade tip can be determined. This angle can be used as a measure for deflection, for example.

The leaf tear can also be evaluated. In particular, it is possible to extend the edge detection to the sheet tear. The course of the sheet tear can be recognized on the basis of the recognized edge. In this case, the detected edge usually runs in an arc. By evaluating the sheet and comparing the respective sheets with each other, it is possible to compare the respective dynamic behavior of the sheets and, if necessary, to detect sheet damage. An emergency shutdown can also be initiated based on the detected deflection.

The detected edges can also be evaluated together with weather data. In particular, rotational speed, deflection, vibration and the like can be related to a wind speed, temperature, wind direction and/or the like. Operating data, in particular SCADA data from the wind turbine, can also be evaluated together with the data from the edges. At least the data can be stored together, chronologically assigned to one another.

In the measurement image, the rotor blade can be measured in its vertical projection, in particular after edge detection. The edge detection can take place at least in part in accordance with the edge detection described above.

In particular, the outline of the rotor blade projected onto the image plane can be measured. It is possible here, for example, for the at least locally maximum chord length of the projected image of the rotor blade to be determined. there is it is possible to determine the number of pixels that run along the straight line of the at least locally maximum long chord.

The actual projected chord length can be determined by defining a path that is represented by exactly one pixel. In this case, for example, a distance determined from a data sheet of a wind turbine can be marked in the image. It is then determined how many pixels are in this marked distance and each individual pixel can then be assigned an actual dimension. In particular, for example, the width of the gondola in the image can be measured. The measure of the actual width of the nacelle can be seen from the data sheet of the wind turbine. A length measure can then be assigned to each individual pixel. This measure of length can be applied to the pixels of the particular chord and determine the actual measure of the length of the projected limb. The angle of attack of a rotor blade can be determined by trigonometric evaluation based on the determined chord length. The distance x of the blade chord (the widest part of the rotor blade) from the blade root can be known. The projection of the blade chords into the camera plane enables the length of the projected blade chord to be determined. With this data the angle of attack can be calculated

A rotor blade has, for example, a first inner radius where it is suspended from the nacelle or the spinner. The radial distance from the pivot point to the end of the rotor blade may be a second radius. The angle of attack α is calculated using L as the measured projected chord length, R1 as the inner radius of the rotor blade and R2 as the outer radius of the rotor blade $$\mathrm{a}=\arcsin \left(\mathrm{L}-\mathrm{R}1/\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="imgf0001.png"\; state="\{\{state\}\}">R</figure-callout>}2\right).$$

The blade pitch angle can be determined individually for each rotor blade. For this purpose, a measurement image can be determined for each rotor blade and the chord length can then be measured as described.

The blade pitch angle of one rotor blade can then be compared with the blade pitch angles of the other rotor blades. Deviations can be used to issue an alarm signal, particularly to a SCADA system.

It is also possible to compare the measured blade pitch angles with the blade pitch angles reported by the wind turbine and measured by sensors. This makes it possible to calibrate the sensors of the wind turbine.

Fig. 1a

eine Seitenansicht einer Windkraftanlage;

Fig.1b

eine Frontansicht einer Windkraftanlage;

Fig. 2

ein erfasstes Bild einer Windkraftanlage;

Fig. 3a

ein Bildausschnitt einer Windkraftanlage mit durchgeführter Kantenerkennung;

Fig. 3b

ein Detail eines Bildausschnitts der Fig. 3a mit einer Helligkeitsverteilung an einer Kante;

Fig. 4a

ein Bild einer Windkraftanlage mit einem initialen Kennpunkt;

Fig. 4b

ein Bild einer Windkraftanlage, bei dem der Kennpunkt von dem initialen Kennpunkt verschoben ist;

Fig. 4c

ein Bild einer Windkraftanlage, bei dem der Kennpunkt von dem initialen Kennpunkt verschoben ist;

Fig. 5

ein zeitlicher Verlauf einer Verschiebung eines Kennpunkts in einer Richtung;

Fig. 6

ein Bild einer Windkraftanlage mit einem definierten Bildbereich;

Fig. 7

ein Bild einer Windkraftanlage, bei dem ein Rotorblatt einen Nulldurchgang hat;

Fig. 8

ein zeitlicher Verlauf eines Mittelwertes von Grauwerten in einem Bildbereich gemäß der Fig. 6;

Fig. 9

ein Bild einer Windkraftanlage mit Kantenerkennung und einer bestimmten zumindest lokal maximalen Sehnenlänge;

Fig. 10

ein Bild einer Windkraftanlage zur Veranschaulichung der Berechnung des Blattanstellwinkels;

Fig. 11

ein schematische Ansicht zur Bestimmung einer Rotation des Spinners.

The object is explained in more detail below with reference to a drawing showing exemplary embodiments. Show in the drawing:

Fig. 1a

a side view of a wind turbine;

Fig.1b

a front view of a wind turbine;

2

a captured image of a wind turbine;

Figure 3a

an image section of a wind turbine with edge detection carried out;

Figure 3b

a detail of a section of the image Figure 3a with a brightness distribution at an edge;

Figure 4a

an image of a wind turbine with an initial characteristic point;

Figure 4b

an image of a wind turbine in which the characteristic point is shifted from the initial characteristic point;

Figure 4c

an image of a wind turbine in which the characteristic point is shifted from the initial characteristic point;

figure 5

a time course of a shift of a characteristic point in one direction;

6

an image of a wind turbine with a defined image area;

7

an image of a wind turbine in which a rotor blade has a zero crossing;

8

a time profile of a mean value of gray values in an image area according to 6 ;

9

an image of a wind turbine with edge detection and a specific at least locally maximum chord length;

10

a picture of a wind turbine to illustrate the calculation of the blade pitch;

11

a schematic view for determining a rotation of the spinner.

Fig. 1a shows a wind power plant 2, also called a wind energy plant or wind turbine, with a tower 4, a nacelle 6, a spinner 8 rotatably suspended on the nacelle 6 and rotor blades 10a-c arranged on the spinner 8. The rotor plane of the rotor blades 10a-c can be slightly inclined with respect to the tower axis 12. A rotor can comprise at least one shaft and the rotor blades 10a-c arranged thereon. The rotor can connect the rotor blades to the generator and can be rotatably mounted.

The tower 4 of the wind turbine 2 is usually fixed to the ground with a foundation (not shown). The tower axis 12 usually runs in the Essentially perpendicular to the ground. The tower axis 12 preferably runs through the tower 4 and the nacelle 6 .

Images from the ground, in particular images of the spinner 8 and the rotor blades 10a-c, are recorded with a camera 14 for an objective determination of operating parameters of the wind power plant 2 . The camera 14 can be a commercially available video camera, in particular a digital, preferably high-resolution video camera. The camera 14 is attached to the tower 4 or to the ground in such a way that it is stationary during the recording and preferably does not vibrate relevantly. The camera 14 is aligned in such a way that its optical axis 16 runs essentially parallel to the tower axis 12 . The field of view 18 of the camera 14 is selected in such a way that the images captured by the camera 14 capture at least parts of the nacelle 6, the spinner 8 and the rotor blades 10a-c.

In the Fig. 1a it can be seen that the camera 14 is installed at the foot of the tower 4 and its optical axis 16 is directed in the direction of the gondola 6 or the spinner 8 .

The present method enables parameters to be determined even if the optical axis 16 does not run parallel to the tower axis 12, but at an acute angle thereto, as in FIG Fig. 1b is shown.

In the Fig. 1b is the wind turbine 2 according to the Fig. 1a shown. A camera 14 is again shown, but this is set up offset to the tower axis 12 at the base of the tower 4 . In this case, the optical axis 16 of the camera 14 is slightly inclined to the tower axis 12. It may be sufficient if the field of view 18 of the camera 14 is such that at least parts of the nacelle 6, the spinner 8 and the rotor blades 10a-c are captured become.

Also according to the example Fig. 1a the optical axis 16 of the camera 14 can run slightly inclined to the tower axis 12. Preferably, the optical axis 16 has an acute angle to the vertical.

2 shows an image 20 which was recorded by the camera 14 . The camera 14 preferably captures moving images. The image 20 is a frame from a sequence of moving images.

Individual images 20 of the moving images are evaluated to determine the operating parameters. The spinner 8, the rotor blades 10a-c and the nacelle 6 as well as a part of the tower 4 are recorded in the image 20. This is made possible by suitably selecting the field of view 18 of the camera 14, in particular by adjusting the focal length, and the optical axis 16 of the camera 14.

In order to determine operating parameters, for example, a vibration of the tower 4 can first be determined and possibly compensated for. For this it is necessary to determine the local displacement of the tower 4 over time. This is preferably possible by observing a prominent point of the wind turbine 2 over time and determining its local shift in the X and Y directions, in particular in the individual images of the sequence, ie also in the image 20 . For this purpose, an initial characteristic point 22 in the image 20 can be determined by a user, for example.

In addition, the image 20 can be subjected to edge detection, so that edges in the image 20 are displayed in a white tone and flat areas are displayed in black. In particular, the image can be transformed into a frequency range and then the spatial frequency of the individual pixels can be used to determine an edge. The sharper an edge in the image, the higher its spatial frequency. For further evaluation, a white value can then be assigned to the pixel depending on the spatial frequency.

In the Figure 3a has been subjected to an edge detection in a section of the image 20 . It can be seen that an edge profile, for example of the spinner 8, is represented by white pixels.

An outer edge is preferably recognized for edge recognition. For edge detection, for example, starting from the pixel of the initial characteristic point, a color value, e.g. a gray value of vertically adjacent pixels (i.e. arranged in a column) can be determined at a distance of n (n preferably <=6) vertically adjacent pixels and a mean value can be calculated from this . Then the color values of pixels in the column of the initial characteristic point are compared with the mean value. The first pixel in this column, whose color value is above the mean value, is preferably recognized as the outer edge and thus as the point for the curve to be approximated. Starting from this pixel, the method described is carried out for a number of adjacent columns, preferably in both directions.

This detected edge profile is used in order to thereby approximate a mathematically determined curve. The curve that comes closest to edge profile 26 is determined.

The calculated curve profile 24 is projected onto the image 20 for the user and it can be seen that the calculated curve profile 24 corresponds approximately to the edge profile 26 of the edge detection.

Since the calculated edge profile 24 follows a mathematically determined curve, its zero point can be determined mathematically. The initial characteristic point 22 is shifted to this zero point.

Figure 3b shows an example of an edge profile after edge detection has been carried out. It can be seen that the white levels fluctuate around the detected edge and the white level flattens out as the distance from the detected edge increases.

The edge detection as used for the Figure 3a described can be performed for each frame. The initial characteristic point 22 is known. Because the position of the camera is stationary relative to the base of the tower 4, the initial reference point 22 in the image section of the image 20 always remains at the same place.

Figure 4a shows the image 20, in which the initial characteristic point 22 was determined. In a subsequent image, such as that shown in Figure 4b is shown, the spinner 8 is shifted to the right and up. With the help of the recognition of the characteristic point, as in connection with the 3 described, a current characteristic point 28 of the spinner 8 is determined. Again, edge detection and curve approximation are used here. The local displacement of the current characteristic point 28 from the initial characteristic point 22 can then be determined. the inside Figure 4b The spinner 8 shown has a displacement 30a in the X direction and a displacement 30b in the Y direction.

Another picture 20 is in the Figure 4c shown. In this image 20, the current characteristic point 28 is shifted from the initial characteristic point 22 by a distance 30c in the X direction and 30d in the Y direction.

The displacement of the current characteristic points 28 from the initial characteristic point 22 can be determined for all images in a sequence of images. It is then possible, as in the figure 5 shown to apply the displacement of the current characteristic point 28 both in the X direction and in the Y direction over the individual images. figure 5 shows an exemplary course of the displacement of the current characteristic point 28 from the initial characteristic point 22 in the Y direction. The same can be done for the X direction. From the course it can be seen that at certain times the current characteristic point deviates far beyond the average of the displacement in the Y direction. This can already be an indication that the wind turbine is not being operated optimally.

In order to further refine the evaluation and to be able to determine further operating parameters, the displacement of the current characteristic point 28 from the initial characteristic point 22 can be compensated for each individual image. For this purpose, the positions of all pixels of an image 20 are shifted by the previously determined displacement of the characteristic point in the X direction and compensates for the previously determined displacement of the characteristic point in the Y direction. This means that the pixels of the image 20 according to the Figure 4b are shifted downwards in the Y direction by the length 30b and leftward in the X direction by the length 30a. The initial characteristic point 22 then lies on the current characteristic point 28. In this consideration, it is assumed that the orientation of the axis of rotation is constant. If there is a change in the axis of rotation (yaw angle) due to a change in the wind direction, this can be recognized, for example, by the curve parameters of the approximated edge changing. A change in the curve parameters can be evaluated, for example, in order to detect a change in the yaw angle. Such a change can also be compensated for, for example, by first changing the approximated curve with regard to the curve parameters that have occurred as a result of the change in the yaw angle. Thus, a curve is assumed which would have been determined had there been no change in yaw rate. Starting from this curve, the characteristic point can then be shifted as described. The order of compensation can also be reversed.

The same applies to the image 20 according to the Figure 4c . Here, for example, all pixels are shifted by the distance 30c to the right in the X-direction and by the distance 30d up in the Y-direction. This also causes the current characteristic point 28 to correspond locally to the initial characteristic point 22 . All images shifted in this way are saved again and can be used for subsequent analysis.

The image progression of the individual images 20 is now free from movements of the tower and/or the gondola and/or the spinner 8 caused by imbalances or wind. The relative movement of the reference point to the position of the camera is compensated. The position of the spinner 8 is identical in all images of a sequence.

With such a motion-compensated sequence of frames, it is possible to determine the period of the rotor. For this purpose, an image area 30 can be determined manually, for example become, as in the 6 shown. The image area 30 preferably overlaps with the central axis 8a of the spinner 8 and is preferably in the area of the gondola 6 or the tower 4, as can be seen in the image 20. The image area 30 is preferably selected in such a way that each individual rotor blade 10a-c crosses the image area in a zero crossing. A zero crossing of a rotor blade 10a-c can be defined, for example, in such a way that the longitudinal axis of a rotor blade 10a-c forms a smallest angle to the tower axis 12, preferably running parallel to the tower axis 12. Furthermore, a tip of this rotor blade is pointing in the direction of the ground or tower at this moment.

Each individual image can then be evaluated with regard to a mean value of a color value, in particular the gray value of all pixels in the image area 30 . Preferably, the mean value of the image area 30 varies when one of the rotor blades 10a-c has a zero crossing. Such a zero crossing is an example in the 7 shown. It can be seen that the moment the rotor blade 10a-c crosses the image area 30, the brightness of the pixels in the image area 30 increases. This is due to the increased reflection of the ambient light by the rotor blade 10a-c compared to the reflection of the ambient light by the underside of the nacelle 6 or the tower 4.

The mean 34 is for all frames, as in the 8 shown, calculated, recorded and preferably also plotted. It can be seen that the mean value 34 deviates greatly from its median at an almost regular distance.

A limit value 32 can be determined. The mean value 34 is compared to the limit value 32 for each frame. If the mean value 34 exceeds the limit value 32, it can be concluded that this image is the image at which a rotor blade 10a-c has a zero crossing. In the 8 each sheet-through image is represented by a mean value peak 34 .

Based on the evaluation of the mean values 34 and the comparison with the limit value 32, those images are determined which have a high probability of zero crossings represent. These can be displayed to the user as sheet traversal images. The user can confirm that the detected image is actually a sheet-through image. In addition, the user can be enabled to look at temporally adjacent preceding or following images of the determined image in the image sequence and to decide whether the determined image is actually the leaf passage image or whether an image is before or after, in particular one or two images before or after , better represent leaf passage. In this case the user can override the automatic determination and manually determine sheet passage images.

After the sheet-pass images have been determined, the distance between the sheet-pass images can be determined in terms of time. Here it is possible, for example, to determine the time or the number of individual images between two sheet passage images. With the help of this time it is possible to determine the period of rotation of the spinner 8. For this it is only necessary to know the number of rotor blades 10a-c. The distance between two blade passage images multiplied by the number of rotor blades gives the period of the spinner 8, i.e. the time it takes the rotor to complete a complete revolution.

The currently determined period is preferably used in each case in order to subsequently determine that image at which the rotor blade 10a-c, which had the zero crossing, is in a horizontal position. Starting from the detected zero crossing, the spinner rotates exactly a quarter of a turn until the rotor blade that just passed through the zero crossing is in a nearly horizontal position. A quarter of the time or the number of frames in a current period corresponds to the time or the number of frames until the rotor blade is in a horizontal position.

This rotor blade is again in a horizontal position at 3/4 of the period. However, it has turned out that the determination of the blade pitch angle is particularly easy when the trailing edge of a rotor blade 10a-c is considered. This The trailing edge can be viewed when the rotor blade 10a-c has completed a quarter turn from the zero crossing.

After the time or the number of frames has been determined, which the rotor blade needs to reach the horizontal position, that frame is determined which is removed from the run frame according to the time or the number of frames.

This then determined image is evaluated in such a way that a maximum chord length of a rotor blade is determined. Edge detection of the image can be used again for this purpose.

Starting from the spinner 8, the chord lengths of the projection of the rotor blade onto the image plane are evaluated in the X direction and the at least locally maximum chord length is determined. This is in the 9 shown. It can be seen that, starting from the spinner 8 or the spinner axis 8a, the distance between an upper rotor blade edge 10a' and a lower rotor blade edge 10a" varies in the X direction. The maximum chord length 40 is determined. This maximum chord length 40 corresponds to the projection of the maximum chord of the rotor blade 10a in the image plane.

With the help of the projection of the blade chord into the image plane and the length of the chord determined in this way, it is possible to calculate the blade angle of attack.

For this purpose, the number of pixels that corresponds to the determined maximum chord length 40 is first determined. A reference distance is then determined in the image 20, for example a diameter 42 of the spinner 8. This diameter 42 of the spinner 8 also has a number of pixels. A measure of a length can be assigned to the number of pixels of the diameter 42 from the data sheet of the wind turbine 2 . This allows each individual pixel to be assigned an actual dimension. Using the number of pixels for the chord length 40 and the one single Pixel associated measure, it is possible to determine the actual measure of the chord length 40.

bestimmen.Subsequently, as in the 10 is shown, from the dimension (L) of the chord length 40 determined in this way, the blade pitch angle α is determined. The inner radius R1 of the rotor blade 10a is known from the data sheet of the wind turbine. Furthermore, the outer radius R2 of the rotor blade 10a is known. Together with the measure of the chord length 40, the blade pitch angle α can be calculated $$\mathrm{a}=\arcsin \left(\mathrm{L}-\mathrm{R}1/\mathrm{<figure-callout\; id="2"\; label="R"\; filenames="imgf0001.png"\; state="\{\{state\}\}">R</figure-callout>}2\right)$$

determine.

The blade pitch angle α can then be determined for each of the rotor blades 10a-c by determining the image of the horizontal position for each of the rotor blades and determining the chord lengths in these images. The blade pitch angles determined for the individual rotor blades can then be compared with one another. As a rule, the blade pitch angles α of all rotor blades 10a-c would have to be approximately the same size, allowing for a tolerance. If a blade pitch angle α of one of the rotor blades 10a deviates from the other rotor blades 10b-c, the wind turbine may be incorrectly adjusted. This can result, for example, from the fact that the sensor of wind turbine 2, which reports the blade pitch angle α of wind turbine 2 or its SCADA, is defective or incorrectly calibrated. A readjustment makes it possible to adjust the blade pitch angles α of all rotor blades 10a-c as equally as possible in order to balance the rotation of the spinner 8.

11 shows a detail of an image of the spinner 8. The vertical image axis y and horizontal image axis x are drawn.

Furthermore, an axis of symmetry 44 through the spinner 8 is shown. It can be seen that the axis of symmetry 44 is inclined by the angle β relative to the vertical image axis y.

In order to be able to determine this angle, various methods are conceivable. A conceivable variant is shown. Starting from the spinner tip, represented by the characteristic point 22, a point 46a,b is first determined on both sides, which is at the same distance from the characteristic point along the curve 24.

It is then determined what distance the respective point 46a, b has from a point 48a, b when the curve 24 intersects with a horizontal line 50. This distance is a measure of the rotation of the spinner 8 relative to the vertical image axis y by the angle β.

Claims (13)

1. Method for determining an operating parameter of a wind power plant comprising:

   - acquiring a sequence of images (20) of the wind turbine (2), in particular during production operation of the wind turbine (2), wherein the images (20) are acquired from below a nacelle (6) of the wind turbine (2) and an optical axis (16) of the image acquisition is parallel or at an acute angle to a tower axis of the wind turbine (2), and

   - determining the operating parameter on the basis of the acquired images,
   characterized in
   that an initial first characteristic point (22) is determined in one of the images (20), the initial first characteristic point (22) being a pixel of the wind turbine (2) whose image position does not change due to the rotation of rotor blades of the wind turbine (2) between two of the images (20),

   - that edge detection is performed in the image (20),

   - that a curve is approximated by the detected edge and thus an approximated curve (24) is determined,

   - that a second characteristic point is determined in the approximated curve (24), the second characteristic point being a point of the approximated curve which is closest to the initial first characteristic point (22), or in that the second characteristic point is a zero point of the approximated curve (24),

   - that the initial first characteristic point (22) is shifted to the second characteristic point and

   - that in temporally successive images a further characteristic point (28) is determined, and in that a spatial position of the further characteristic point (28) is compared with the spatial position of the shifted initial first characteristic point (22), and that a spatial deviation of the further characteristic point (28) from the shifted initial first characteristic point (22) is determined, the spatial deviation representing the operating parameter, the operating parameter being the deflection of the tower and/or of the nacelle.
2. Method according to claim 1,
   characterized in
   that at least parts of the wind turbine (2) whose position is substantially independent of the rotation of a rotor, in particular parts of the spinner (8) and/or the nacelle (6) and at least parts of at least one rotor blade of the wind turbine (2), are detected in the field of view of the image acquisition, in particular that at least one spinner tip of the spinner (8) is detected in the field of view of the image acquisition.
3. Method according to claim 1 or 2,
   characterized in
   that moving images of the wind turbine (2) are acquired and/or in that individual images of the acquired images are evaluated.
4. Method according to claim 1,
   characterized in
   that the approximated curve (24) is a course of a conical section.
5. Method according to one of the preceding claims,
   characterized in
   that one of the images (20) is shifted depending on the determined deviation of the further characteristic point (28) in such a way that the deviation is substantially compensated.
6. Method according to one of the preceding claims,
   characterized in
   that an image section is determined, that a color value, in particular a gray value of the image section, in particular a mean color value of the image section is determined, and that those images (20) are determined in which the color value exceeds a limit value and are determined as blade passage images.
7. Method according to claim 6,
   characterized in
   the image section is determined in an area between the spinner (8) and the tower of the wind turbine (2).
8. Method according to one of claims 6 or 7,
   characterized in
   the time and/or the number of images (20) is determined which lie between each two blade passage images.
9. Method according to claim 8,
   characterized in
   that a period of a complete rotation of the rotor blades is determined depending on the determined time and/or the determined number of images (20), wherein the period is determined in particular N times the determined time and/or the determined number of images (20), with N corresponding to the number of rotor blades.
10. Method according to claim 9,
    characterized in
    that, as a function of the period, a time and/or a number of images (20) is determined in which a rotor blade has performed a quarter turn, and in that a measurement image is determined therewith.
11. Method according to claim 10,
    characterized in
    that a maximum chord length of a detected rotor blade is determined in the measurement image.
12. Method according to claim 11,
    characterized in
    that the blade pitch angle of the rotor blade is determined from the determined chord length.
13. Method according to claim 12,
    characterized in
    in that the blade pitch angle is determined for each rotor blade of the wind turbine (2) and/or in that the determined blade pitch angles of the rotor blades are compared with one another and/or in that, in the event of a deviation of the blade pitch angles of two rotor blades greater than a limit value, an error signal is output and/or in that, in the event of a deviation of the blade pitch angles of two rotor blades greater than a limit value, maintenance of the wind turbine is initiated.

Images